The UK government conducted a six-month trial of Microsoft 365 Copilot across 300 civil servants in late 2025. The headline result: 26 minutes saved per day. The detail: the savings were concentrated in administrative tasks — meeting summarization, email drafting, document formatting — and vanished almost entirely for analytical and decision-making work. HMRC participants rated the tool 3.2 out of 5 on "would recommend." A separate DWP trial found similar results. That is not a product category in hypergrowth. That is a product finding its floor.

Microsoft has been disciplined about not disclosing Copilot adoption or revenue separately. The data that leaks — and it always leaks — suggests 365 Copilot seats are concentrated in a small number of large enterprises that are running pilots, not broad deployments. The average seat count per Copilot customer is estimated by several sell-side analysts at under 500. For context, Microsoft 365 has roughly 450 million paid commercial seats. Even the most optimistic extrapolation puts Copilot penetration at single digits. As of Q4 FY2026, the company disclosed over 30 million paid Copilot seats, up from 20 million the prior quarter — accelerating but still under 7% of the M365 base.

The pricing doesn't help. At $30 per user per month, Copilot adds roughly 30-50% to the average 365 enterprise license cost. For a 10,000-seat deployment, that's $3.6 million annually. The ROI case for "26 minutes saved per day of administrative time" is not $3.6 million. At a fully loaded cost of $75 per hour, 26 minutes is roughly $32.50 per day in saved labor cost. Over 250 working days, that's about $8,100 per year. The Copilot license costs $360. The ROI is there on paper — but it assumes every saved minute translates to productive output, which anyone who has worked in an office knows is a fantasy.

The real issue is not whether Copilot is useful. It is. Meeting summaries alone are worth something. The issue is whether the market has priced Microsoft for Copilot being a $20 billion annual business or a $5 billion annual business. At current multiples, the implied expectation is closer to the former. Microsoft trades at roughly 33x forward earnings — a premium to the S&P that reflects, in part, an AI revenue growth story that is notably light on disclosed data.

Here's a heuristic: if a product is genuinely transformative, the company selling it publishes granular usage metrics. Salesforce publishes Agentforce conversation counts. Shopify publishes GMV by channel. Microsoft publishes "Copilot customers increased 60% quarter-over-quarter" — a percentage growth number without a base, which means nothing. The opacity is the signal.

There is a counterargument worth taking seriously. Copilot is a platform play, not a point product. It integrates across the entire Office suite, and the integration value compounds as Microsoft adds agents, connectors, and Copilot Studio for customization. The 26 minutes will become 46 and then 66 as the product improves — and the improvement curve for LLM-based products is steeper than traditional software. This is reasonable. It is also unproven.

What is proven: the UK trial is not an outlier. Multiple enterprise IT leaders I've spoken with report similar patterns — Copilot is popular with the subset of users who live in email and documents, ignored by engineers and analysts, and not yet adopted broadly enough to show up in departmental productivity metrics. The product is fine. The gap between "fine" and "priced for transformation" is where the disappointment lives. The market has not yet had the conversation about which side of that gap Copilot is on.